SHORT COURSE

		   "PATTERN RECOGNITION"

ABSTRACT:

This course gives a short introduction into the field of
automatic pattern recognition, with a special
emphasis of pattern classification. First the
motivation and the necessary terms are given,
together with three illustrative examples. Then the
groundwork is laid how to approach classification in
a Bayesian probabilistic framework, introducing
error, rejection and risk for decision making. The
principal part deals with the different classifier
architectures, model-based parametric Bayesian
classifiers, Parzen windows, Nearest-Neighbor
classifiers, Linear Discriminant Functions with one
representative learning algorithm, the delta rule,
which is extended to the generalized delta rule in the
multilayer perceptron, radial basis function networks
and polynomial classifiers.
Dimensionality reduction is presented as an
important preprocessing step in pattern recognition,
consisting of feature selection and feature extraction.
Feature selection search strategies are outlined
together with the selection criteria. Feature extraction
comprises Principal Component Analysis and Linear
Discriminant Analysis. The topic of error estimation
is used to measure the performance of a classifier.
The Sammon map is used to visualize high-dimensional
data. The text closes with the c-means clustering
algorithm as a representative of unsupervised learning.

A practical presentation of the theoretical concepts is
included, using the 'TOOLDIAG' pattern recognition software.

---------------------------------------------------------------------------

			INDEX

1. MOTIVATION AND TERMINOLOGY
 Introduction
 Patterns, Pattern Recognition, Features and Classification 
 Constraints 
	 Pattern classification and higher-level information processing 
	 Numerical features vs. symbolic features 
	 Probabilistic framework 
	 Time stationary patterns vs. dynamic patterns 
 Basic Model of a Pattern Recognition System 
 Nomenclature 
 Examples of Pattern Generating Processes 
	 Example "Iris": Iris Flowers 
	 Example "Reactor": Fault detection in a chemical reactor 
	 Example "Vision": Object recognition in an industrial assembly task 
 Further Reading 
 Software Support 

2. PROBABILISTIC FRAMEWORK FOR PATTERN CLASSIFICATION 
 Bayes Theorem 
 Bayes Decision Rule For Minimum Error Classification 
 Bayes Decision Rule For Maximum Likelihood Classification 
 Bayes Decision Rule For Minimum Risk Classification 
 Inference and Decision 
 Model Assumption and Parameter Estimation
	 Model assumption 
	 Parameter estimation 
	 Assuming multivariate Gaussian distribution and estimating
			its parameters 

3. CLASSIFIER MODELS, SUPERVISED LEARNING AND PROBLEMS 
 Model-based Parametric Classifiers 
 Parzen Windows 
 The 1-Nearest Neighbor Classifier 
 The K-Nearest Neighbor Classifier 
 The Nearest Prototype Classifier 
 Linear Discriminant Functions for Classification 
	 The Model 
	 Gradient Descent Learning and Least Mean Squared Error 
 Generalized Linear Discriminant Functions for Classification 
	 Universal Function Approximators 
 	Curse of Dimensionality and the Bias-Variance Dilemma 
 Polynomial Classifier 
 Radial Basis Function Networks 
 Multilayer Perceptron 
	 Architecture 
	 Delta Rule and Generalized Delta Rule 

4. DIMENSIONALITY REDUCTION BY FEATURE SELECTION AND EXTRACTION 
 Feature Selection
	 Selection criteria 
	 Search strategies 
	 Limitations of Feature Selection 
 Feature Extraction 

5. ERROR ESTIMATION

6. HIGH-DIMENSIONAL DATA VISUALIZATION 

7. UNSUPERVISED LEARNING 

BIBLIOGRAPHY